If your business serves customers through a chatbot, whether on your website, on WhatsApp or over the phone, that chatbot has to make it clear that it is an AI. Article 50 of the EU AI Act has required this since 2 August 2026, unless it is already obvious. A single line at the start of the conversation is usually enough. This guide is for owners and managers of small and medium-sized businesses: what the rule says, when you do not need a notice, and notice examples ready to copy.
What Article 50 requires and since when
Regulation (EU) 2024/1689, known as the EU AI Act, devotes its Article 50 to transparency. Its first paragraph requires AI systems that talk directly to people to be designed so that those people are "informed that they are interacting with an AI system".
The rule is EU law, but it does not stop at EU borders. Article 2 also covers providers based outside the EU when they place AI systems on the EU market or when the system's output is used in the EU. So if your chatbot serves people in the EU, it is sensible to assume this applies to you, wherever your business is based.
The exception sits in the same sentence. You do not need a notice when it is "obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect", taking into account the circumstances and context of use. Paragraph 5 adds how to do it: in a clear and distinguishable manner, and "at the latest at the time of the first interaction".
These obligations have applied since 2 August 2026, the general date of application set by Article 113 of the Regulation. Article 50 also deals with other cases, such as deepfakes or emotion recognition, which are not the chatbot notice and fall outside this guide.
Wasn't the AI Act postponed? If you have read that, the detail matters. Regulation (EU) 2026/1744, known as the Digital Omnibus on AI, mainly delays the obligations for high-risk systems. For Article 50, according to the timeline summarised by Adequa (in Spanish), there is only a transition period until 2 December 2026. It concerns the technical marking of synthetic content from systems already on the market. The chatbot notice has not moved.
Even if you have read that the AI Act was postponed, your chatbot notice has applied since 2 August 2026.
To interpret the rule, the reference point is the European Commission's Guidelines on transparency obligations, published on 20 July 2026. They are not binding: they explain how the Commission reads the article. The practical criteria in this guide come from them, as summarised by law firms such as McCann FitzGerald and DLA Piper. In Spain, AESIA, the Spanish Agency for the Supervision of Artificial Intelligence, summarises on its website the situations Article 50 covers.
The Regulation provides for fines for breaching Article 50 and, for SMEs, the lower of the two amounts set in its Article 99 applies. But the reason to add the notice is not the fine: it is that your customer knows who they are talking to.
Who it affects and when you do not need a notice
Article 50(1) is not addressed to whoever uses the chatbot, but to the provider. Under Article 3 of the Regulation, a provider is whoever develops an AI system, or has one developed, and places it on the market or puts it into service under its own name or trademark. For a small business, that leaves two common situations.
If the chatbot carries your brand
If the assistant appears under your business name, you will normally be treated as the provider. That is usually the case even if it runs on an AI model from OpenAI or Anthropic, or an agency built it for you. This follows from the definition in Article 3, which includes whoever has the system developed by someone else. In practice, it is up to you to make sure the notice exists and is visible.
If you use another company's chatbot as it is
If you install a third-party chat window without changing its branding, the provider will normally be that company and the notice should come built in. Your part is not to switch it off, hide it or cover it with your own welcome message. If you have customised it so much that it looks like yours, it is worth checking which side you fall on.
When it is obvious and when it is not
According to the Commission's Guidelines, the "obvious" exception is read narrowly. It is measured against the real audience using the chatbot, not against someone who knows about technology. People knowing that chatbots exist does not mean they recognise one in every conversation.
- Usually not obvious: a customer service chatbot on your website or on WhatsApp, or an assistant that answers phone calls.
- May be obvious: a coding assistant for developers or an internal tool used by trained staff.
If your chatbot serves customers, start from the assumption that it needs a notice. Borderline cases are for an adviser to assess with your situation in front of them, not for a blog post.
Where and when it has to appear
The Regulation asks for a notice that is clear, distinguishable and delivered at the latest at the first interaction. According to the Commission's Guidelines, that translates into these points:
- At the start. Before or at the beginning of the conversation, not after three replies.
- Inside the conversation. Tucking it into your terms and conditions does not work. Neither does a vague label such as "assistant" or a generic banner along the lines of "this site uses AI".
- In words that leave no doubt. Say "AI" or "artificial intelligence".
- In the channel's format. It can be text, audio, visual or a combination, so on a call it is fine to say it out loud.
- Accessible. Screen readers should read it, and older people, children and people with disabilities should understand it.
- With reminders where needed. One notice at the start is usually enough, but in sensitive conversations, such as health advice or vulnerable people, it is worth repeating.
If you run the same assistant on several channels, think about each one separately. A customer who messages you on WhatsApp has not seen your website's welcome message.
Notice examples for web, WhatsApp and phone
A notice that works has three parts: which business the assistant belongs to, that it is an AI, and how to reach a person. Article 50 does not require the third, but it saves frustration for anyone who would rather be helped by someone on your team.
A good notice fits in one line: whose it is, that it is an AI and how to reach a person.
The examples are illustrative and use a fictional business, Example Clinic. Change the name, what the assistant does and how to reach a person to suit your case. For each channel there is a short version and one with a handover to a person.
Website chat
Put the notice in the welcome message, which appears before the customer types anything. Back it up with a fixed label next to the assistant's name in the chat header.
Hi, I'm Example Clinic's AI assistant. How can I help?
Hi, I'm an artificial intelligence assistant for Example Clinic. I can help with appointments and opening hours; if you'd rather talk to a person, tap "Talk to the team".
Header label: AI assistant · Example Clinic.
On WhatsApp, the most direct approach is to put the notice in the assistant's first message, before it answers the question. If the customer writes again days later, repeating it in one line bothers no one.
Hi, this is Example Clinic's AI assistant. Tell me what you need.
Hi, I'm Example Clinic's artificial intelligence assistant. If at any point you want to talk to a person, type PERSON and someone from the team will pick up.
In another guide we will cover how to connect WhatsApp to n8n, an automation tool, through the official API, which is the connection WhatsApp offers businesses. We will also dedicate one to the types of chatbot WhatsApp Business allows.
Phone and voice assistants
On a call, the notice is spoken as soon as the call is answered, before asking for any details. Keep it short, in a clear voice, with an easy way through to a person.
Hello, you've reached Example Clinic. You're speaking with an artificial intelligence assistant.
Hello, you've reached Example Clinic and you're speaking with an artificial intelligence assistant. If you'd rather talk to a person, say "person" or press zero.
Common mistakes when writing it
Most notices that fail do so not out of bad intent, but because they sit where nobody looks or use words that say nothing.
- Hiding it. In the website footer, in the privacy policy or in text that only appears after a click.
- A human name with no notice. "I'm Laura, how can I help?" is misleading. The name can stay if the notice comes with it: "I'm Laura, Example Clinic's AI assistant".
- Only giving notice on one channel. Getting the website chat right and forgetting WhatsApp or the phone.
- Denying it is a bot when asked. Some law firms, such as DLA Piper, read this as an obligation in the Guidelines. Either way, it is good practice for the assistant always to tell the truth.
- Leaving no way out. A notice with no way to reach a person may meet the minimum, but it leaves the customer stuck.
- Confusing it with the privacy notice. Article 50(6) makes clear that it does not replace other transparency obligations, such as those under the GDPR (General Data Protection Regulation). They are two separate notices and you need both.
How to build the notice into an n8n chatbot
n8n is a tool for building workflows that connect apps and run tasks on their own. It is our go-to tool for 24/7 customer service and sales chatbots.
In the chatbots we build, the notice is visible from the very first moment: in the assistant's name, in a short line of text inside the chat window next to the button to talk to the AI, or in both places. That way people know they are talking to an AI before they type anything.
The underlying idea is that the notice should not depend on the AI model remembering to say it. These are the parts of the workflow where it fits:
- First reply node. Each step in an n8n workflow is a node. The first one that replies sends a fixed message with the notice, before the model answers.
- One variable per channel. The notice text for web, WhatsApp or voice is stored in one place, so changing it does not mean touching the whole workflow.
- An instruction in the prompt. The prompt, the instructions the model receives, tells it to acknowledge that it is an AI if asked.
- Handover to a person. A keyword or a button stops the assistant and alerts the team so they can take over the conversation.
- A log. The workflow records that the notice was shown, with date and channel, in case you ever need to check.
If the workflow will run on your own server and you are unsure where to host your n8n, we cover it when comparing shared hosting and a VPS (virtual private server). And if your chatbot is already live and you are not sure its notice is right, we can review it for you.
What to check in your chatbot
- The notice appears before or in the first message, on every channel where the assistant responds.
- It says "AI" or "artificial intelligence", not just "assistant" or "virtual assistant".
- It is visible inside the conversation, not only in the legal pages.
- It includes your business name and an easy way to reach a person.
- If the chatbot belongs to another company, its notice is still switched on and visible.
- On the website, screen readers can read it; on the phone, it is said when the call is answered.
- Sensitive conversations include a reminder.
- The assistant does not deny being an AI when asked.
- The AI notice and the privacy notice are both there, each in its place.
In summary
Since 2 August 2026, a chatbot that serves people in the EU has to make clear that it is an AI, unless that is obvious. The notice goes at the start of the conversation, on every channel and in plain words. If the chatbot carries your brand, it will normally be up to you to make sure it gives notice.
Our view: the notice does not scare customers away. What does is finding out afterwards that you were talking to a bot.